Daniil Belyaev is an entrepreneur and the founder of Syntra Systems, an IT company from Tashkent. We spoke to him not about how many apps his team has built, but about what is happening to technology right now: AI agents, the fate of apps and websites, the future of professions, deepfakes and levels of trust in artificial intelligence.
Key takeaways
The interview was published in the Syntra Systems blog on 15 August 2026. The answers appear as they were given: we changed only the layout and added editorial notes where the subject has a primary source or a separate blog article.
The first part of the conversation is about what has changed over the past couple of years: AI is no longer only someone to talk to, it starts doing things.
I think the second is gradually turning into the first. For the first couple of years we mostly talked to AI: write me a letter, make an image, explain this topic. Now the next stage is starting — it does not only answer, it acts.
You say: find me a good hotel in Almaty for the weekend, compare the options and pick the best one. In time you will not have to open ten sites, read reviews and fill in forms yourself. That is far more interesting than another image generator.
Technically that is not the hard part. It gets much more interesting when it has access to your calendar, email, banking, documents and work systems. Then it becomes a digital assistant: “I have a meeting at ten tomorrow, prepare the client information, look through the previous correspondence and remind me of the important questions.”
But a question appears here: how ready we are to give AI access to our lives. An assistant that knows nothing about you is useless. An assistant that knows everything is a little frightening.
Next come interfaces: what a person actually uses when they need a taxi, a ticket or an answer about a company.
This is one of the most interesting questions. Today the habit is: need a taxi — one app, music — another, a ticket — a third. The interface may gradually change: instead of working out which app to use, a person simply says what they need and the AI chooses the service.
Apps become infrastructure under the bonnet, and the main interface becomes conversation. You used to look for the “Pay” button. Later you will simply say: “pay”.
They are not going anywhere tomorrow. What changes is how people arrive at them. The internet used to start with a search engine: you typed a query, got ten links and went off to read. Now people increasingly ask AI first and get a ready answer.
For business that is a serious shift: being in the results is no longer enough. AI has to know who you are, what you do and why you are worth recommending. In effect a new kind of optimisation appears — no longer only for the search engine, but for the model as well.
The third part is about work: which tasks go to the machine and what stays with people.
Yes, but there is a lot of misreading here. AI removes individual tasks, not whole professions. If a person spends eight hours moving data from one spreadsheet to another, I see no tragedy in a machine taking that over. The profession does not necessarily disappear — they may start doing the work they never had time for.
Some professions will change a great deal, though. Above all those with a lot of repetitive intellectual work: basic analytics, content, support, translation, document processing.
Programmers are not going anywhere. But the programmer of the future and the programmer of five years ago are different people. You used to be able to say: I can write code. Today that is not enough.
AI can write a fairly large chunk of a program, so a person’s value shifts from “I can write code” to “I understand what needs to be built”. The irony is that the better the AI programmer becomes, the more valuable the human who understands the business and can frame the task properly.
It is a useful skill, but I would not overrate it. In time AI will understand ordinary human language well enough. It matters far more to understand what you want. If a person does not know what they want, a perfect prompt will not help — it is like giving a Ferrari to someone who does not know where to drive.
Independence is a subject of its own: the more access and tools AI gets, the more the limits matter.
It is already starting to, and that is both the most interesting and the most dangerous trend. Agents can use tools, look for information, run programs, work with files and perform chains of actions. The more independence, the more potential mistakes. So the question is no longer only “how clever is the AI”, but “how well can we control it”.
If there is, I hope at least without traffic jams. Seriously though, the realistic danger today is not a robot that decided to destroy humanity, but a person who gave AI too many rights and set no limits.
Imagine an employee allowed to read all the corporate mail, write to clients, spend company money and delete files. Now imagine they work around the clock and never sleep. That is roughly why AI safety is becoming a large industry of its own.
Editor’s note In Uzbekistan the limits for AI-based systems are already written into law: Law No. ZRU-1115 of 21 January 2026 added article 7-1 to the law on informatisation. Information resources created with artificial intelligence, and systems running on it, must not harm a person, their life, health, freedom, honour and dignity, or violate their other inalienable rights.
The next part is about AI moving beyond the screen: into robotics, cars and wearables.
That is what I am waiting for most. Robots are harder than AI on a computer: AI can make a mistake in a text and you get a funny letter. A robot makes mistakes physically, and that gets expensive.
But AI plus robotics is a genuinely interesting combination: AI learns to understand the world around it, and the robot gains the ability to act on it. That is why AI is actively moving from screens into robots, cars, devices and wearables.
I would not bet money on that in the coming years. But the phone may stop being the main way we interact with technology. AI glasses, devices with constant voice interaction and new wearables are already appearing.
The idea is simple: why take out the phone, open an app and press five buttons every time, if you can just say what you need. The phone gradually turns from the main interface into one of the interfaces.
Here the conversation comes down to trust: where a person will hand the decision to a machine, and where they want to confirm it themselves.
Not everything. People like control. Nobody wants to wake up and find that AI has bought them a ticket to Ulaanbaatar because it decided that was the perfect holiday destination.
So the future, I think, is not about full autonomy but about levels of trust. “Find the options” — you can do that yourself. “Pick the best one” — that can be delegated. “Buy it for a hundred dollars” — probably too. “Spend ten thousand” — there you want to hear a human “are you sure?”.
That is a real problem. We are used to treating a photograph or a video as proof that something happened. That era is ending. Checking the source, not only the content, is becoming more important.
I think fairly soon we will trust not the photograph itself but the digital signature, the origin of the file and a confirmed source. The question used to be “is this true?”. Soon a second one will be added: “how do we know it is true?”.
Editor’s note Checking where a file came from is no longer an idea but a standard. The C2PA consortium has published the Content Credentials specifications: a signed history of how the file was created and edited travels with it, and that history can be verified by software.
The closing part is about what surprises the founder himself and what he advises people outside IT.
Not a particular technology, but the speed of change. A big technology used to appear once every few years. Now you can open the news in the morning and see something that seemed like science fiction a month ago. Not long ago AI was mostly a chatbot; now we discuss agents, AI in devices and robotics. Even inside the industry you have to keep relearning.
Not try to compete with AI where it is stronger. Better to learn to work with it. You do not have to become a programmer or a tech geek — you have to learn to use technology as a tool. The way everyone once learned to use search, a smartphone and a banking app. Only now, instead of “find me information”, it becomes “do this for me”. That is the habit we all still have to acquire.
Smart fridges that send a notification saying I have run out of eggs. I know perfectly well without artificial intelligence that I have run out of eggs. Now if the fridge orders them, pays and meets the courier — then we can talk.
The conversation looks a few years ahead, but some of its subjects became ordinary working tasks long ago. Below is where we covered them in detail.
| Subject from the conversation | Where it is covered in detail |
|---|---|
| An assistant that answers customers | AI assistants and chatbots for business |
| AI access to mail, documents and work systems | AI security in a company |
| A website that models have to understand | GEO optimisation: getting into ChatGPT and Gemini answers |
| Measuring mentions of a company in AI answers | The AI visibility report |
At Syntra Systems we build such solutions around specific processes: an assistant that answers customers from your own documents, processing of incoming requests and paperwork, a model connected to the CRM and to access rules. The work starts not with the model but with a review of the processes.
The nearest entry point on the service page is an AI audit of processes for $400: a list of processes for AI, an automation map and an implementation plan with priorities.
Let’s discuss your project
Tell us what you need, and we will estimate the timeline and cost and suggest a solution.
Daniil Belyaev is an entrepreneur and the founder of Syntra Systems, an IT company from Tashkent and an IT Park resident. The company works on AI solutions and automation, CRM and ERP, websites and platforms, Telegram bots, mobile apps and cybersecurity. The interview was published in the company blog on 15 August 2026.
A chatbot answers with text, while an agent performs actions: it uses tools, looks for information, works with files and carries a chain of steps through to a result. Hence the difference in risk: a bot’s mistake is a poor answer, an agent’s mistake is an action already taken that has to be rolled back.
It is a rule for deciding which actions AI performs itself and which a person confirms. In the interview it sounds like this: “Find the options” — you can do that yourself. “Pick the best one” — that can be delegated. “Buy it for a hundred dollars” — probably too. “Spend ten thousand” — there you want to hear a human “are you sure?”.
It is more reliable to check the origin rather than the picture itself. The C2PA consortium has defined the Content Credentials format: a signed history of how the file was created and edited travels with it. Such data is verified by software, unlike attempts to spot a fake by eye.
Start by making sure the website carries clear answers to customer questions: what you do, for whom, what it costs and how the work is organised. Then watch what the models actually say about the company and measure it regularly. This is covered in detail in the articles on GEO optimisation and the AI visibility report.
Roughly the same as an employee with broad rights: a separate account, the minimum necessary access, a log of actions and human confirmation for anything involving money, mailings or deleting data. What exactly to check is covered in the article on AI security in a company.
